The intelligibility of speech in elementary school classrooms
J. S. Bradleya兲
National Research Council, Montreal Road, Ottawa, K1A 0R6, Canada
H. Sato
Institute for Human Science & Biomedical Engineering, National Institute of Advanced Industrial Science
and Technology (AIST), 1-1-1 Higashi, Tsukuba, Ibaraki 305-8566, JapanII
共Received 16 July 2007; revised 11 November 2007; accepted 8 January 2008兲
This is the second of two papers describing the results of acoustical measurements and speech
intelligibility tests in elementary school classrooms. The intelligibility tests were performed in 41
classrooms in 12 different schools evenly divided among grades 1, 3, and 6 students 共nominally 6,
8, and 11 year olds兲. Speech intelligibility tests were carried out on classes of students seated at their
own desks in their regular classrooms. Mean intelligibility scores were significantly related to
signal-to-noise ratios and to the grade of the students. While the results are different than those from
some previous laboratory studies that included less realistic conditions, they agree with previous
in-classroom experiments. The results indicate that +15 dB signal-to-noise ratio is not adequate for
the youngest children. By combining the speech intelligibility test results with measurements of
speech and noise levels during actual teaching situations, estimates of the fraction of students
experiencing near-ideal acoustical conditions were made. The results are used as a basis for
estimating ideal acoustical criteria for elementary school classrooms. 关DOI: 10.1121/1.2839285兴
PACS number共s兲: 43.55.Hy, 43.71.Gv 关RYL兴
I. INTRODUCTION
There has recently been renewed interest in achieving
improved acoustical conditions in classrooms. Although
there is a general consensus as to acoustical criteria for good
classrooms,1 the supporting evidence from studies in actual
classrooms is limited. The renewed interest in classroom
acoustics is related to our growing understanding of the
negative effects of ambient noise and poor room acoustics on
children’s ability to learn in schools 共see also Anderson2 for
an extensive review兲.
There is evidence that increased levels of noise affect
memory3 and are associated with decreased reading
scores.4–6 It is reasonable to assume that a mechanism to
explain the association of decreased educational progress,
such as the effect of noise on reading scores, is simply that
noise interferes with verbal communication 共especially that
between teachers and students兲, which is the predominant
mechanism for learning in elementary schools.
It is well known that younger children have greater difficulty understanding speech in even modest levels of ambient noise.7–9 In fact several authors have reported results
showing that the ability to recognize speech in noise improves systematically with age.7,10,11 Although it is clear that
children need quieter conditions and corresponding larger
signal-to-noise ratios than adults to achieve high speech recognition scores,8 and that the younger the children, the quieter the conditions should be, the results of the various previous studies reveal large differences and do not agree well
with previous results of tests carried out in actual
classrooms.12
a兲
Electronic mail: [email protected]
2078
J. Acoust. Soc. Am. 123 共4兲, April 2008
Pages: 2078–2086
Figure 1 compares results from Marshal11 and Elliott7
with previous classroom measurements of speech intelligibility scores versus A-weighted speech–noise level differences,
S / N共A兲. Marshal and Elliott’s results are both based on
simple word recognition tasks and both indicate large variations in intelligibility scores with the age of the listener.
However, when examined in detail, these two sets of results
show quite different effects of age and show some much
lower intelligibility scores than the previous classroom results for 12 year olds. The previous classroom results were
from a 1986 study that included ten classes of 12 year old
students12 and used a rhyme test in which subjects identified
the correct response of several possible rhyming words. The
subjects were seated in their regular seats in their own classrooms and listened to the recorded speech material from a
small loudspeaker with directionality similar to a human
talker.
The differences in Fig. 1 are thought to be due to the
laboratory studies using monaural headphone presentation of
speech and noise signals, which increases the negative effects of noise and reverberation. Listening naturally with two
ears leads to a “binaural advantage” that can make it easier to
understand speech in noise. MacKeith and Coles13 measured
binaural advantages equivalent to as much as an 18 dB improvement in signal-to-noise ratio for extreme cases in free
field conditions. Neuman and Hochberg9 tested 5 to 13 year
old children, finding small binaural advantages for all ages in
reverberant conditions. They pointed out that the binaural
advantages would probably be much larger in lower signalto-noise situations. Nábflek and Robinson14 tested subjects
with ages from 10 to 72 years and found an average binaural
advantage corresponding to a 5% increase in intelligibility
scores in reverberant conditions but without significant
masking noise. They did not include the effects of lower
0001-4966/2008/123共4兲/2078/9/$23.00
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
used to estimate the fraction of students in the measured
classrooms who would experience these ideal conditions.
That is, a new basis for classroom acoustics criteria as well
as estimates of the negative impact of common existing conditions were produced.
100
90
80
Intelligibility, %
70
60
II. MEASUREMENT PROCEDURES
50
40
30
Marshall, age: Elliott, age:
5
7
7
9
9
11
11
13
1986 classrooms
20
10
0
-15
-10
-5
0
5
10
15
20
S/N(A), dB
FIG. 1. Laboratory word intelligibility test results from Marshall—Ref. 11
共5, 7, 9, and 11 year olds兲 and Elliott—Ref. 7 共7, 9, 11, and 13 year olds兲
showing the effect of listener age on intelligibility scores vs A-weighted
speech–noise level differences, S / N共A兲, and compared with previous inclassroom intelligibility test results from 12 year olds 共Ref. 12兲.
signal-to-noise ratios where larger effects would be expected.
It is clear that the older laboratory based speech intelligibility
studies in which monaural headphone presentation was used
would exaggerate the negative effects of noise on speech
recognition but the magnitude of these effects cannot be accurately estimated.
Similarly there are previous studies indicating that
younger children’s ability to understand speech is more adversely affected by room reverberation.9,14,15 Some early
studies used monaural headphone presentation of the test
signals,10 which would exaggerate the negative effects of reverberation on speech recognition scores.16 Because shorter
reverberation times led to improved speech recognition
scores in these tests, they have led some to recommend very
short reverberation times for classrooms. Of course, excessive absorptive treatments may control reverberation but will
also decrease effective speech levels and exaggerate the
more common problem of inadequate signal-to-noise ratios
in classrooms.
The limitations of the various previous studies carried
out in laboratory settings can be avoided by testing children
in their classrooms where the children listen in realistic conditions with both ears. The work reported in this paper and in
a companion paper17 was intended to provide a more complete basis for deriving acoustical criteria for classrooms
from speech intelligibility tests of students in their own
classrooms and acoustical measurements in the same classrooms. More specifically it was intended to show younger
children’s ability to understand speech in real classroom conditions as a function of signal-to-noise ratios, varied room
acoustics, and the age of the children. At the same time it
was planned to determine speech and noise levels in classrooms during actual teaching situations in order to understand the actual signal-to-noise ratios experienced in active
classrooms. By combining these two types of information,
this paper estimates ideal acoustical conditions in classrooms
for children of various ages. At the same time the data are
J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
The measurements included speech recognition tests of
complete classes of grades 1, 3, and 6 students 共nominally 6,
8, and 11 year olds兲 in their own classrooms. The signal-tonoise ratios experienced by the students were varied over a
wide range by varying the speech playback level. Room
acoustics conditions were varied by testing in a number of
different classrooms. Speech and noise levels were measured
during the speech tests as well as during normal teaching
activities. Room acoustics measurements were made with the
classroom occupied and also unoccupied.
A. Speech recognition tests
The speech intelligibility scores were obtained using the
Word Identification by Picture Identification 共WIPI兲
test11,18,19 that includes four lists of 25 phonetically balanced
simple nouns. This was selected as an easy test that 6 year
olds and older students could quickly learn and respond to
individually in a classroom situation. It consists of simple
test words said to be familiar to 5 year olds and these were
presented in the carrier phrase, “Please mark the _ _ _ _
now.” While sitting at their desks in their regular classroom,
the students responded by placing a sticker on one of six
pictures to indicate the correct word. Each sentence was approximately 3 s long and the next sentence was played to the
students when all were ready to proceed.
The tests were carried out in 41 classrooms evenly distributed among grade 1, grade 3, and grade 6 students in 12
different schools. The schools were in relatively quiet rural
areas and small towns in Eastern Ontario, Canada. A total of
840 students were evaluated in 41 classrooms. Grade 1 students were each tested at two different signal-to-noise values
and the other students at three different signal-to-noise values. A total of 2285 individual speech recognition tests were
obtained.
All students in each class with parental permission participated in the tests. The parental permission form asked
whether the student had any known hearing problems. Almost all students with known hearing impairment did not
receive parental permission and did not participate. As a result the test results of the few students with reported hearing
problems were not included in these analyses because of the
very small numbers of these students.
The speech source was a small loudspeaker with similar
directionality to that of a human talker. Digital recordings of
the WIPI test material were made in an anechoic room so
that they were reflection free and with negligible noise. A
female talker was used and the recordings were edited to use
exactly the same version of the carrier phrase for all test
words and to have the same sound levels for all test words.
Varied S / N共A兲 values were obtained by changing the playback level of the speech material relative to the existing
J. S. Bradley and H. Sato: Speech in classrooms
2079
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
natural ambient noise in the classroom. This included some
tests with 20– 30 dB S / N共A兲 values to determine intelligibility scores in truly ideal conditions for these students. This
made it possible to obtain speech recognition scores for a
very wide range of signal-to-noise ratios with natural ambient noises. Room acoustics conditions were varied by carrying out the tests in a number of different classrooms. It was
not possible to artificially modify room acoustics conditions
as part of the experiments.
B. Acoustical measurements
Speech and noise levels were recorded during the speech
recognition tests at four positions in the fully occupied classrooms. There were on average 20.4 students in each classroom 共s.d.⫾ 3.96兲 and hence there were about five students
near each measuring microphone. These recordings were
used to determine speech and noise levels during the tests by
statistical analysis of the distributions of recorded sound
levels.17
Speech and noise levels were also recorded at the same
four microphone positions during a normal teaching activity
in each classroom. Recordings were made during a period
when the teacher planned to be mostly talking to the class of
students as a complete group. Finally, ambient noise levels
were also recorded at these same four positions when the
classroom was unoccupied.
Room acoustics parameters were also measured from
impulse responses obtained at the same microphone locations for both occupied and unoccupied conditions. These
included decay times and energy ratios as described in the
companion paper17 along with the various speech and noise
level measurements. All sounds were digitally recorded on a
portable computer. To avoid running cables through the
classrooms, the signals from the four microphones were connected to a central computer via 16 bit digital transmitters
and receivers.
III. MEAN TRENDS OF SPEECH RECOGNITION TEST
RESULTS
FIG. 2. Mean intelligibility scores of groups of approximately five students
vs A-weighted speech–noise level difference, S / N共A兲. The error bars indicate the standard deviations of the scores of each group of students.
There is also a large amount of scatter about the mean
trends that tends to increase with decreasing S / N共A兲 values.
The scatter is partly due to the approximately five students
near each microphone being slightly different distances from
the microphone. The larger scatter at lower S / N共A兲 values
may be indicative of how students react to more difficult
listening conditions. At lower S / N共A兲, some students can
still do quite well, but others may more or less give up and
get much lower scores.
The best-fit regression lines to the data in Fig. 2 are
repeated in Fig. 3 and are compared with the previous speech
intelligibility scores from 12 year olds in classrooms. The
previous results12 used a rhyme test with simple rhyming
words presented in a carrier phrase. Given the amount of
scatter in both sets of data, and possible larger inaccuracies
in the older results, there is remarkable agreement between
the old results for 12 year olds and the results for 11 year
olds 共i.e., the grade 6 students兲. This agreement suggests that
A. Speech recognition scores as a function of S / N„A…
2080
J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
90
80
70
Intelligibility, %
Speech recognition scores were first examined as a function of S / N共A兲 as this was the key independent variable that
was varied in these experiments. The speech recognition
scores were averaged over the scores of the approximately
five students located close to each microphone in each classroom. These average scores are plotted versus the measured
signal-to-noise ratios 关S / N共A兲兴 in Fig. 2. They are plotted
versus S / N共A兲 separately for the results of the grades 1, 3,
and 6 students. An analysis of variance of the scores showed
that there were highly significant main effects of age and
S / N共A兲 共p ⬍ 0.001兲 as well as a highly significant interaction
effect of these two independent variables, age and S / N共A兲
共p ⬍ 0.001兲. That is, although there is significant scatter in
the results, there are highly significant effects related to
S / N共A兲 and the age of the listeners. The younger children
need significantly higher S / N共A兲 values to obtain the same
intelligibility scores as the older children in these classrooms.
100
60
50
40
30
Grade 6
Grade 3
Grade 1
1986 Schools
20
10
0
-10
-5
0
5
10
15
S/N(A), dB
FIG. 3. Comparison of the best-fit regression lines from Fig. 2 with the
mean trend of previous speech intelligibility scores for 12 year olds in
classrooms 共Ref. 12兲.
J. S. Bradley and H. Sato: Speech in classrooms
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
TABLE I. Summary of mean values and standard deviations 共␴兲 of key
room-acoustics parameters for occupied conditions. C50, EDT, and T60 are
midfrequency results for the combined 500 and 1000 Hz octave bands.
S / N共A兲 is the signal-to-noise ratio in terms of A-weighted speech–noise
level differences.
100
Intelligibility, %
95%
90
8.5 dB
grade 6
12.5 dB 15.5 dB
grade 3 grade 1
80
Grade 6
Grade 3
Grade 1
70
0
5
10
15
20
Grade 1
Mean
␴
Grade 3
Mean
␴
Grade 6
Mean
␴
C50共500– 1000兲
EDT共500– 1000兲
T60共500– 1000兲
S / N共A兲
8.9
2.19
0.40
0.10
0.43
0.09
6.5
7.30
8.6
2.41
0.42
0.12
0.44
0.11
3.8
5.2
8.8
2.94
0.37
0.13
0.40
0.10
3.4
9.0
S/N(A), dB
FIG. 4. 共Color online兲 Expanded view of the mean trends from Fig. 2
showing the point at which each regression line reaches a 95% speech
intelligibility score.
the results are a more representative indication of children’s
ability to understand speech as a function of S / N共A兲 in real
classroom conditions than some previous laboratory speech
tests.
Figure 4 shows an expanded view of the best-fit regression lines from Fig. 2. The performance of the three age
groups can be compared by considering the required S / N共A兲
for each group to achieve near-ideal conditions for speech
communication. For the results of simple word intelligibility
tests such as the WIPI test, a speech intelligibility score of
95% correct is used to indicate near-ideal conditions, because 95% correct scores are readily achievable in high
S / N共A兲 conditions. For example, Fig. 2 shows that for very
high S / N共A兲 共+23 to +30 dB兲, the grade 1 and 3 students
scored ⬃98% correct and the grade 6 students scored
⬃99.5% correct. That is, although the younger children
might be expected to find the test a little more difficult, all
three age groups are capable of getting higher scores than
95% in very high S / N共A兲 conditions.
The mean trends in Fig. 4 show that the grade 6 students
could, on average, achieve 95% correct scores for a S / N共A兲
of +8.5 dB. However, the grade 3 students required
+12.5 dB S / N共A兲 and the grade 1 students required
+15.5 dB S / N共A兲 to obtain a mean score of 95% correct. In
this case there is a 7 dB difference between the average
needs of grade 1 and grade 6 students. That is, the grade 1
students would need a 7 dB greater S / N共A兲 value, or a corresponding 7 dB quieter ambient noise level, to obtain the
same intelligibility scores as the grade 6 students. Further, it
is likely that the grade 6 students would have somewhat
lower speech intelligibility scores than would young adult
listeners in the same situations.9
frequency values of early-to-late arriving sound ratios 共C50兲,
early decay times 共EDT兲, reverberation times 共T60兲, and
A-weighted signal-to-noise ratios 关S / N共A兲兴. The acquisition
of these data are described in the companion paper.17 It was
hoped that there would be sufficient variation of room acoustics conditions among the classrooms to determine the additional effects of room acoustics on speech intelligibility
scores. In practice, Table I indicates relatively small variations about the mean conditions that were close to ideal.
To investigate possible additional effects of room acoustics on intelligibility scores, multiple regression analyses
were performed regressing speech intelligibility scores on
values of S / N共A兲, S / N共A兲2, and one of the room acoustics
parameters. Table II summarizes the results in terms of the
resulting R2 共coefficient of determination兲 values. The R2
values for the combination of the S / N共A兲 and S / N共A兲2 are
first given for each grade level of students. If there are significant additional effects of room acoustics, then when values of one of the room acoustics parameters were added to
the regression analysis, the R2 value would be expected to
increase. For the grade 1 results none of the room acoustics
parameters added significantly to the prediction and the R2
values did not increase. For the grade 3 and 6 results, adding
one of the room acoustics parameters to the prediction did
result in modest but significant increases in the prediction
accuracy of the intelligibility scores. Of the three room
acoustics parameters considered, C50 values tended to be
slightly more effective in increasing the R2 values.
Figure 5 illustrates the resulting multiple regression
equations for combinations of S / N共A兲, S / N共A兲2, and C50
TABLE II. R2 values from multiple regression analyses of intelligibility
scores on the predictors shown at the top of each column. The subscript ns,
indicates that the room acoustics predictor variable 共C50, EDT, or T60兲 did
not add significantly to the prediction 共p ⬍ 0.05兲.
N
S / N共A兲
S / N共A兲2
S / N共A兲
S / N共A兲2
C50
S / N共A兲
S / N共A兲2
EDT
S / N共A兲
S / N共A兲2
T60
136
96
156
0.707
0.625
0.355
0 . 710ns
0.660
0.508
0 . 707ns
0.655
0.475
0 . 707ns
0.663
0.474
B. Effects of room acoustics on speech recognition
scores
Table I lists the mean values of key room-acoustics
parameters20 along with the standard deviations of these parameters for each grade level group. These included midJ. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
Grade 1
Grade 3
Grade 6
J. S. Bradley and H. Sato: Speech in classrooms
2081
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
TABLE III. Average intelligibility scores for each grade and overall average
standard deviations 共␴兲 for all grades determined for each S / N共A兲 interval.
Speech intelligibility, %
100
80
S / N共A兲 共dB兲 Avg. grade 1共%兲 Avg. grade 3共%兲 Avg. grade 6共%兲 ␴共%兲
60
−10
−5
0
5
10
15
20
25
40
C50 = 6 dB
C50 = 9 dB
C50 = 12 dB
20
(a) grade 3
Speech intelligbility, %
0
100
80
38.0
56.5
73.2
84.7
91.3
94.8
96.4
97.3
¯
69.0
81.5
89.3
93.6
95.9
97.0
¯
70.5
80.5
87.9
92.6
95.0
97.3
98.3
¯
16.50
12.30
9.00
5.68
3.50
2.20
1.29
0.70
60
Canadian schools, these results suggest that room acoustics
in these elementary school classrooms tend to be homogeneously reasonably acceptable.
40
C50= 6 dB
C50 = 9 dB
C50 = 12 dB
20
(b) grade 6
0
-15
-10
-5
0
5
10
15
S/N(A), dBA
FIG. 5. Multiple regression results for prediction of speech intelligibility
scores from S / N共A兲, S / N共A兲2, and C50 values for 共a兲 grade 3 results and 共b兲
grade 6 results. Results are given for C50 values of 6, 9, and 12 dB roughly
corresponding to the range of frequently found conditions in the measured
classrooms.
values for the grade 3 and grade 6 results. The regression
equations in Fig. 5 are given by the following:
Grade 3,
SI = 2.495 S/N共A兲 − 0.110 S/N共A兲2
+ 0.98C50 + 70.37,
Grade 6,
SI = 0.772 S/N共A兲 − 0.0189 S/N共A兲2
+ 1.53C50 + 74.46.
The above-mentioned grade 3 results suggest that a 1 dB
change in C50 values would result in about a 1.0% change in
intelligibility scores. However, the grade 6 results indicate a
1.53% change in intelligibility scores would result for a 1 dB
change in C50 values. If most of the data are within ⫾1 s.d.
of the mean, most C50 values would be within a range of
about 6 – 12 dB depending on the grade. The largest effect
would be for the grade six results where changes in C50
values are likely to result in changes to intelligibility scores
of up to about 9% 共obtained by multiplying the standard
deviation value in Table I for C50 values in grade 6 classrooms by the coefficient of C50 in the equation for the abovepresented grade 6 results, and doubling the result to include
positive and negative deviations about the mean兲. Of course,
this is much smaller than the effect of S / N共A兲 values illustrated in Fig. 2.
These data do not show large effects of room acoustics
parameters because there were not large variations in room
acoustics conditions among the various classrooms. This is
illustrated in Fig. 5 that plots the regression equations for the
case when C50 was used as the added room acoustics parameter. Similar results were obtained using EDT and T60 values
as the added room acoustics parameter. Since these classrooms are assumed to be “typical” of many classrooms in
2082
J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
IV. DISTRIBUTION OF INTELLIGIBILITY SCORES
ABOUT THE MEAN TRENDS
While Sec. II discusses the mean trends of speech intelligibility scores, it is clear that many individual student results deviate significantly from the mean trends. That is,
while the average student may for some condition be able to
understand reasonably well, many cannot. It is therefore important to also examine the distribution of intelligibility
scores about the mean trend. The ultimate goal is to determine the fraction of the students that can understand speech
well at each S / N共A兲 value and the required conditions to
enable most children to understand speech well.
The distribution of the intelligibility scores about the
mean trends seen in Fig. 2 was analyzed by dividing the
results into 5 dB wide S / N共A兲 intervals for the data from the
students of each grade. It was then possible to examine the
distribution of scores within each of these S / N共A兲 segments.
Although in most cases there were approximately normal
distributions of scores in each segment, in some cases there
were not adequate numbers of data points to provide regular
distributions. A procedure was required to approximate the
distributions in all segments of the data. To do this, the mean
scores and the standard deviations of the scores about the
mean values were calculated for each S / N共A兲 segment.
These values were then plotted versus the mean S / N共A兲 values for each interval. The mean values are listed in Table III
and are different for each grade and follow trends almost
identical to those in Fig. 2.
The standard deviations of the scores in each S / N共A兲
interval were similarly plotted versus the mean S / N共A兲 value
for each interval. As illustrated in Fig. 6, the results for all
three grades follow an approximately similar trend but with
some considerable uncertainty in the lowest S / N共A兲 interval.
The mean trend in Fig. 6 gives a good estimate of the variation in the standard deviation of scores over a wide range of
S / N共A兲 values and is quite adequate for the purposes of
these analyses in which the focus is on very good conditions
for speech communication. Therefore this mean trend is used
as an estimate of the standard deviations for all grades.
关There is some uncertainty in the mean trend at the lowest
S / N共A兲 category where the mean S / N共A兲 is −10 dB. This is
J. S. Bradley and H. Sato: Speech in classrooms
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
0.3
S/N(A)
grade 1
grade 3
grade 6
20 dB
Fraction of responses
standard deviation of SI, %
18
12
6
0.2
15 dB
0.1
10 dB
-5 dB
-10 dB
0 dB
5 dB
0.0
0
0
-15
-10
-5
0
5
10
15
20
20
25
40
60
80
100
Speech intelligibility, %
S/N(A), dB
to be expected as conditions get more difficult and at extremely low S / N共A兲 values intelligibility scores must eventually all approach 0% with a standard deviation of zero.兴
The mean trend standard deviations from Fig. 6 are included
in Table III.
Assuming a normal distribution is a reasonable approximation to the distribution of scores in each S / N共A兲 interval,
one can estimate the distributions of scores for each interval
from the mean and standard deviation of the scores in each
interval. That is, the number of speech intelligibility scores
in each S / N共A兲 interval can be estimated from the following
expression for a normal distribution:21
y = N/兵␴冑2␲其e−共x − ␮兲
2/2␴2
,
共1兲
where N is the total number of intelligibility scores in the
distribution for one S / N共A兲 interval, y is the number of
scores at intelligibility x, x is the speech intelligibility score,
␴ is the standard deviation of the intelligibility scores in each
S / N共A兲 interval, and ␮ is the mean speech intelligibility
score in each S / N共A兲 interval.
With the mean scores and standard deviations for each
S / N共A兲 interval given in Table III, one can construct a
simple mathematical model of the speech intelligibility
scores for the responses of students in each grade. This
model can describe the distribution of scores as well as the
mean trend of the scores for each age of student. Figure 7
shows the resulting speech intelligibility distributions in each
S / N共A兲 interval for the grade 1 results.
One can more easily describe the fraction of students
experiencing some high level of speech intelligibility by replotting the information as cumulative probability plots. This
is done in Fig. 8 for the data from all three grade levels. The
goal is to determine the required S / N共A兲 at each grade level
for students to experience near-ideal conditions. Near-ideal is
again defined as corresponding to speech intelligibility
scores of 95% or better.
Figure 8 can be used to determine the fraction of students at each S / N共A兲 category that would experience nearideal conditions with intelligibility scores of 95% or better.
For the example of the grade 6 students at a 20 dB S / N共A兲,
essentially all would experience 95% intelligibility or better.
J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
FIG. 7. Distribution of the speech intelligibility scores in S / N共A兲 intervals
from −10 to +20 dB for the grade 1 results.
At a S / N共A兲 of +15 dB only 25.4% of the grade 6 students
would not experience 95% or better speech intelligibility.
One might therefore argue that the common recommendation
for a +15 dB S / N共A兲 value12,22 is satisfactory for the grade 6
students because at this S / N共A兲, 74.6% of the students
would experience near-ideal conditions.
For the grade 3 students at a +15 dB S / N共A兲 value, only
54.5% would experience 95% or better speech intelligibility
and for the grade 1 students, only 36.6% would experience
this near-ideal speech intelligibility for conditions of +15 dB
S / N共A兲. However, for a +20 dB S / N共A兲, 74.8% of the grade
1 students would experience 95% or better speech intelligibility. It is therefore important to note that a +15 dB S / N共A兲
does not provide near-ideal conditions for most of the grade
1 students.
V. COMPARISON OF ACTUAL S / N„A… VALUES WITH
IDEAL REQUIREMENTS
A. Ideal S / N„A… goals
The data in Fig. 8 can be replotted as the percentage of
students who would experience 95% speech intelligibility or
1.0
0.8
Grade 6
0.6
0.4
S/N(A)
0.2
Fraction of responses
FIG. 6. Mean trend of standard deviations of speech intelligibility scores for
all three grades as a function of the S / N共A兲 values.
0 dB
-5 dB
20 dB
15 dB
10 dB
5 dB
0.0
0.8
Grade 3
0.6
0.4
S/N(A)
0.2
5 dB
0 dB
-5 dB
20 dB
15 dB
10 dB
0.0
0.8
Grade 1
20 dB
0.6
15 dB
0.4
0.2
0.0
S/N(A)
0 dB
-5 dB
40
60
Intelligibility score, %
5 dB
10 dB
80
100
FIG. 8. 共Color online兲 Cumulative probability distributions of intelligibility
scores by S / N共A兲 category and student grade level. Vertical dashed line
indicates 95% speech intelligibility.
J. S. Bradley and H. Sato: Speech in classrooms
2083
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
for intermediate grade results and perhaps even assume that
grade 7 and 8 results would yield a curve only slightly to the
left of the grade 6 results in Fig. 9.
To accommodate the needs of younger children and/or
to include even a modest safety factor leads to required
S / N共A兲 values greater than the common recommendation for
an ideal S / N共A兲 of +15 dB.
Fraction better than 95% SI
1.0
0.8
Grade1
Grade3
Grade6
0.6
0.4
B. Conditions in active classrooms
0.2
0.0
-10
-5
0
5
10
15
20
As part of this work, speech and noise levels were recorded in active classrooms and separate speech and noise
levels were determined using a statistical procedure.17 共“Active” indicates the classrooms were fully occupied with students active in educational activities with their teacher.兲 The
mean teacher speech level at the microphone locations near
the students was 59.5 dBA with a standard deviation of
⫾5.5 dB. The distribution of measured S / N共A兲 values in
active classrooms is shown in Fig. 12 of Ref. 17. The mean
S / N共A兲 was approximately +11 dB. Although a S / N共A兲 of
+11 dB is often said to correspond to somewhat acceptable
conditions for speech communication, Sec. V A demonstrated that much higher S / N共A兲 values are required for good
speech communication for younger children. In the actual
teaching situations shown in Fig. 12 of Ref. 17, S / N共A兲 values of +15 dB or more only occurred in 6.25% of the measured cases. The S / N共A兲 conditions during the measurements of actual teaching sessions were considerably inferior
to the S / N共A兲 values required for near-ideal conditions in
Sec. IV.
Knowing children’s ability to understand speech at various S / N共A兲 values and knowing the distribution of S / N共A兲
values found in active elementary school classrooms, one
can now estimate the proportion of the students in the measured classrooms who would experience near-ideal conditions for speech communication. First, the distribution of
measured S / N共A兲 values in the active classrooms was recalculated with the same 5 dB intervals as were used in Table
III and Figs. 7 and 8. These values are listed in the left-hand
3 columns of Table IV and show that most of the measured
conditions 共61.6%兲 had S / N共A兲 values in the category with a
mean S / N共A兲 of +10 dB.
The middle three columns in Table IV show the percentage of students who would experience 95% speech intelligi-
25
S/N(A), dB
FIG. 9. Percentage of students who would experience 95% speech intelligibility or higher as a function of the S / N共A兲 value.
better as a function of the existing S / N共A兲 value, as illustrated in Fig. 9. This again assumes that 95% speech intelligibility on a simple word intelligibility test represents nearideal conditions for speech communication. In this format
one can directly determine the S / N共A兲 required for a certain
percentage of the students to experience 95% speech intelligibility or better. That is, one could choose a desired S / N共A兲
goal so that some large percentage of the students would
experience these near-ideal conditions for speech communication. For example, one might aim for acoustical conditions
in which at least 80% of the students would experience 95%
speech intelligibility or better. Figure 9 shows that for the
grade 6 results a S / N共A兲 value of just above +15 dB is required. The grade 3 results indicate a minimum S / N共A兲 of
+18.5 dB would be required and the grade 1 results indicate
a minimum required S / N共A兲 of close to +20.5 dB.
One could require that a larger or smaller percentage of
the students should experience such near-ideal conditions.
For example, requiring that 90% of the students should experience such near-ideal conditions rather than 80% would
increase the minimum S / N共A兲 values needed by about 2 dB.
One might also include a small safety factor to be sure that
the desired S / N共A兲 value is actually achieved in real classrooms. Some differences in approach are possible, but Fig. 9
can serve as the basis for setting acceptable S / N共A兲 criteria
for elementary school students. One could readily interpolate
TABLE IV. Calculation of the percentage of students at each grade level who would experience near-ideal acoustical conditions for speech communication in
the measured active classrooms. Columns 1–3, distribution of S / N共A兲 values while teachers were talking. Columns 4–6, percentages of students who would
experience 95% speech intelligibility or better in each S / N共A兲 category. Columns 7–9, total percentages of students who would experience near-ideal
conditions in the measured active classrooms.
Mean
S / N共A兲 共dB兲
0
5
10
15
20
Total
2084
N
cases
0
9
69
34
0
112
%
cases
0.0
8.0
61.6
30.4
0.0
100.0
Percent scoring ⬎95% SI
%cases⫻ % students
Grade 1
Grade 3
Grade 6
Grade 1
Grade 3
Grade 6
3.9
13.0
36.6
12.0
26.6
54.5
21.4
40.5
74.6
0.31
8.03
11.12
0.96
16.39
16.55
1.72
24.93
22.63
19.5
33.9
49.3
J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
J. S. Bradley and H. Sato: Speech in classrooms
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
TABLE V. Calculation example to determine maximum acceptable ambient
noise levels.
59.5 dB A
−5.5 dB A
−20 dB A
34 dB A
35 dB A
Mean classroom speech level from teachers
Standard deviation of measured speech levels
Required S / N共A兲 for grade 1 students
Required maximum ambient noise level
ANSI S12.60 maximum ambient noise level criteria
bility or better in each of these same S / N共A兲 categories and
these were obtained from Fig. 8. By multiplying the percentage of cases in each S / N共A兲 category by the corresponding
percentage of students who would experience 95% speech
intelligibility or better in that category, the total percentage
of all students who would experience near-ideal conditions is
obtained in the three right-hand columns for each age group.
These final percentages are summed over all three S / N共A兲
categories at the lower right-hand side of Table IV.
The quite surprising result is that only 19.5% of the
grade 1 students would experience near-ideal conditions during the measured teaching activities. This rises to 33.9% of
the grade 3 students and 49.3% of the grade 6 students. This
is in spite of the fact that room acoustics parameters were
near-ideal and the rooms seemed to have only minimal
acoustical problems to the 共adult兲 experimenters carrying out
these tests.
C. Maximum ambient noise level criteria
By combining the various results from this study, one
can estimate maximum acceptable ambient noise levels that
would provide near-ideal speech communication for students
of various ages. Table V illustrates the process for grade 1
students for whom a minimum S / N共A兲 of 20 dB was found
to be necessary in the previous sections of this paper. This
estimate results in a recommended maximum ambient noise
level that is almost identical to that recommended in the
ANSI S12.60 classroom acoustics standard. This is strong
confirmation of the validity of the recommendation in ANSI
S12.60.
Similar estimates could be made for grade 3 and 6 students. If these estimates included a small safety factor of
2 – 4 dB to account for various uncertainties, the results
would again be close to the recommendations of ANSI
S12.60. It is therefore justifiable and practical to recommend
a maximum ambient noise level in all elementary school
classrooms of no more than 35 dBA. It is important to realize too, that although occupied noise levels may be higher
than 35 dBA, lower unoccupied levels lead to lower occupied levels17 and hence it is particularly important to make
unoccupied ambient noise levels as low as possible.
Unfortunately the current data are not adequate to estimate ideal room acoustics criteria for classrooms. There was
simply very little variation in room acoustics conditions.
This means that the conclusions with respect to desirable
S / N共A兲 and ambient noise levels are directly applicable to
classrooms with room acoustics conditions similar to those
in these classrooms. However, the room acoustics measurement results in Table I show that conditions in these classJ. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
rooms were close to most recommended values for classrooms. Thus, the recommended S / N共A兲 values can be said to
be an important component of an ideal classroom.
VI. DISCUSSION AND CONCLUSIONS
This study has provided data that better describe the
abilities of elementary school children to understand speech
in noise in real classrooms of schools near Ottawa, Canada.
The results are better because they are from a large sample of
children and are based on natural binaural listening in actual
classrooms with realistic ambient noises. These results also
realistically include the other distractions that are expected to
occur in actual classrooms such as those from other students
both in the classroom and in adjacent spaces.
The form of the relationships between intelligibility
scores and S / N共A兲 values is similar to previous in-classroom
speech intelligibility tests but different than earlier laboratory
studies using monaural headphone playback of the speech
material.
The mean trends of the results indicate that grade 1 students 共6 year olds兲 require 7 dB higher S / N共A兲 values to
achieve the same speech intelligibility scores as would grade
6 students 共11 year olds兲. Although no adult data were obtained, it is likely that young adult listeners would get somewhat higher speech intelligibility scores than the grade 6
students at the same S / N共A兲 conditions.9
There is also much scatter in the speech intelligibility
scores about the mean trends indicating that many students
would often have more difficulty understanding speech than
indicated by the mean trends. The distribution of speech intelligibility scores about the mean trends was therefore also
examined and a mathematical model of the means and distributions of scores was developed to more completely define
children’s abilities to understand speech in noise.
From this model estimates of the S / N共A兲 values required for grade 1, grade 3, and grade 6 students to experience near-ideal conditions for speech communication were
made. Near-ideal conditions were defined as 95% speech intelligibility scores on simple word intelligibility tests and all
ages of student could do better than this in very high S / N共A兲
conditions. For 80% of the students to experience such nearideal conditions, S / N共A兲 values of +20, +18, and +15 dB
would be required for grade 1, grade 3, and grade 6 students,
respectively.
Measurements during actual teaching activities showed
an average S / N共A兲 of 11 dB and in only 6.25% of the cases
were the S / N共A兲 +15 dB or higher. In the actual teaching
situations only 19.5% of the grade 1, 33.9% of the grade 2,
and 49.3% of the grade 6 students would experience nearideal conditions for speech communication. That is, in the
measured classrooms that appeared to have acceptable
acoustical conditions to the adult experimenters, less than
half of the students would experience near-ideal speech communication.
The inability of younger children to understand many of
the words that a teacher is saying must make it more difficult
for the children to learn new concepts. There is a growing
literature of results indicating that increased noise levels are
J. S. Bradley and H. Sato: Speech in classrooms
2085
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
associated with a number of educational factors such as delayed reading ability, effects on memory, and student
behavior.2
Further calculations based on the new measurements led
to estimates of maximum acceptable ambient noise levels
that were very close to the 35 dBA recommendation in
ANSI S12.60.
The range of room acoustics conditions measured in the
classrooms was quite small and close to values thought to be
optimum. Although there were significant effects of room
acoustics parameters, the limited range of the data made it
impossible to produce new estimates of ideal room acoustics
conditions for speech communication in classrooms as a
function of student age. Further research is required to consider the question of optimum room acoustics criteria to
maximize intelligibility and the quality of speech communication.
ACKNOWLEDGMENTS
This work was supported by a grant from the Canadian
Language and Literacy Research Network. The very helpful
cooperation of the teachers and administration of the Upper
Canada District School Board made this work possible.
1
American National Standards Institute, “Acoustical performance criteria,
design requirements, and guidelines for schools,” ANSI S12.60, New
York, 2002.
2
K. Anderson, “The problem of classroom acoustics: The typical classroom
soundscape is a barrier to learning,” Semin. Hear. 25, 117–129 共2004兲.
3
S. Hygge, “Classroom experiments on the effects of different noise
sources and sound levels on long-term recall and recognition in children,”
Appl. Cognit. Psychol. 17, 895–914 共2003兲.
4
A. L. Bronzaft and D. P. McCarthy, “The effect of elevated train noise on
reading ability,” Environ. Behav. 7, 517–528 共1975兲.
5
A. L. Bronzaft, “The effect of a noise abatement program on reading
ability,” J. Environ. Psychol. 1, 215–222 共1981兲.
2086
J. Acoust. Soc. Am., Vol. 123, No. 4, April 2008
6
G. W. Evans and L. Maxwell, “Chronic noise exposure and reading deficits,” Environ. Behav. 29, 638–656 共1997兲.
7
L. L. Elliott, S. Connors, E. Kille, S. Levin, K. Bell, and D. Katz, “Children’s understanding of monosyllabic nouns in quiet and in noise,” J.
Acoust. Soc. Am. 66, 12–21 共1979兲.
8
L. L. Elliott, “Performance of children aged 9 to 17 years on a test of
speech intelligibility in noise using sentence material—With controlled
word predictability,” J. Acoust. Soc. Am. 66, 651–653 共1979兲.
9
A. Neuman and I. Hochberg, “Children’s perception of speech in reverberation,” J. Acoust. Soc. Am. 73, 2145–2149 共1983兲.
10
T. Finitzo-Hieber and T. W. Tillman, “Room acoustics effects on monosyllabic word discrimination ability for normal and hearing-impaired children,” J. Speech Hear. Res. 21, 440–458 共1978兲.
11
N. B. Marshall, “The effects of different signal-to-noise ratios on the
speech recognition scores of children,” Ph.D. thesis, University of Alabama, Tuscaloosa, AL, 1987.
12
J. S. Bradley, “Speech intelligibility studies in classrooms,” J. Acoust.
Soc. Am. 80, 846–854 共1986兲.
13
N. W. MacKeith and R. R. A. Coles, “Binaural advantages in hearing of
speech,” J. Laryngol. Otol. 85, 213–232 共1971兲.
14
A. K. Nábflek and P. K. Robinson, “Monaural and binaural speech perception in reverberation for listeners of various ages,” J. Acoust. Soc. Am.
71, 1242–1248 共1982兲.
15
A. K. Nábělek and J. M. Pickett, “Reception of consonants in a classroom
as affected by monaural and binaural listening, noise, reverberation and
hearing aids,” J. Acoust. Soc. Am. 56, 628–639 共1974兲.
16
J. P. Moncur and D. Dirks, “Binaural and monaural speech intelligibility in
reverberation,” J. Speech Hear. Res. 10, 186–195 共1967兲.
17
H. Sato and J. S. Bradley, “Evaluation of acoustical conditions for speech
communication in active elementary school classrooms,” J. Acoust. Soc.
Am. 123, 2064–2077 共2008兲.
18
M. Ross and J. Lerman, “A picture identification test for hearing impaired
children,” J. Speech Hear. Res. 13, 44–53 共1970兲.
19
M. Ross and J. Lerman, Word Intelligibility by Picture Identification 共Stanwix House, Pittsburgh, 1971兲.
20
International Organisation for Standardisation, “Acoustics—Measurement
of the reverberation time of rooms with reference to other acoustical parameters,” ISO3382, Geneva, Switzerland.
21
G. F. Ferguson and Y. Takane, Statistical Analysis in Psychology and
Education, 6th ed. 共McGraw Hill, New York, 1989兲.
22
T. Houtgast, “The effect if ambient noise on speech intelligibility in classrooms,” Appl. Acoust. 14, 15–25 共1981兲.
J. S. Bradley and H. Sato: Speech in classrooms
Downloaded 21 Jun 2011 to 150.29.215.199. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp